gRNAde: Geometric Deep Learning for 3D RNA inverse design.

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• RNA sequence design often focuses on a single structure, neglecting 3D conformational diversity.
• Existing computational methods may not fully capture RNA structure-dynamics relationships.

Purpose of the Study:

• Introduce gRNAde, a geometric RNA design pipeline for sequence design based on 3D backbone structures.
• Explicitly account for RNA structure and dynamics in sequence design.
• Evaluate gRNAde's performance against established tools like Rosetta.

Main Methods:

• Utilizes a multi-state Graph Neural Network and autoregressive decoding.
• Generates RNA sequences conditioned on one or more 3D backbone structures.
• Operates on 3D RNA backbones with unknown base identities.

Main Results:

• Achieved higher native sequence recovery rates (56%) on a fixed backbone benchmark compared to Rosetta (45%).
• Demonstrated utility in multi-state design for flexible RNAs and fitness landscape analysis.
• Showcased a 50% success rate in designing pseudoknotted RNA structures, surpassing Rosetta's 35%.

Conclusions:

• gRNAde offers a significant advance in computational RNA sequence design by incorporating 3D structural and dynamic information.
• The pipeline is faster and more accurate than existing methods for various RNA design tasks.
• Experimental validation confirms gRNAde's effectiveness, particularly for challenging structures like pseudoknots.